Ecosystems (2010) 13: 1097–1111 DOI: 10.1007/s10021-010-9376-8 2010 Springer Science+Business Media, LLC

Ecosystem Carbon Storage Across the Grassland–Forest Transition in the High of Manu National Park,

Adam Gibbon,1* Miles R. Silman,2 Yadvinder Malhi,1 Joshua B. Fisher,1 Patrick Meir,3 Michael Zimmermann,3 Greta C. Dargie,3 William R. Farfan,2,4 and Karina C. Garcia2,4

1Environmental Change Institute, School of Geography and the Environment, University of Oxford, Oxford OX1 3QY, UK; 2Department of Biology, Wake Forest University, 1834 Wake Forest Rd., Winston Salem, North Carolina 27106, USA; 3School of Geosciences, University of Edinburgh, Edinburgh EH8 9XP, UK; 4Universidad Nacional de San Antonio Abad, Cusco, Peru

ABSTRACT Improved management of carbon storage by ter- aboveground carbon decreased from 15.8 on the restrial biomes has significant value for mitigating puna to 8.6 in the transition zone and 2.1 in the forest. climate change. The carbon value of such manage- No significant relationships were found between ment has the potential to provide additional income carbon densities and slope, altitude or fire disturbance to rural communities and provide biodiversity and history, though grazing (for puna) was found to re- climate adaptation co-benefits. Here, we quantify duce aboveground carbon densities significantly. We the carbon stores in a 49,300-ha landscape centered scaled our study sites to the study region with remote on the –grassland transition of the high sensing observations from Landsat. The carbon Andes in Manu National Park, Peru. Aboveground sequestration potential of improved grazing man- carbon densities were measured across the land- agement and assisted upslope treeline migration was scape by field sampling of 70 sites above and below also estimated. Afforestation of puna at the treeline the treeline. The forest near the treeline contained could generate revenues of US $1,374 per ha over the 63.4 ± 5.2 Mg C ha-1 aboveground, with an addi- project lifetime via commercialization of the carbon tional 13.9 ± 2.8 Mg C ha-1 estimated to be stored credits from gains in aboveground carbon stocks. in the coarse roots, using a root to shoot ratio of Uncertainties in the fate of the large soil carbon stocks 0.26. Puna grasslands near the treeline were found under an afforestation scenario exist. to store 7.5 ± 0.7 Mg C ha-1 in aboveground bio- mass. Comparing our result to soil data gathered by Key words: Peru; Manu National Park; treeline; Zimmermann and others (Ecosystems 13:62–74, puna; upper tropical montane cloud forest; carbon 2010), we found the ratio of belowground: stocks.

Received 25 October 2009; accepted 12 August 2010; published online 8 September 2010 INTRODUCTION Author Contributions: AG, MRS, YM, JBF, and PM conceived of or It is now widely recognized that anthropogenic designed this study; AG, MRS, MZ, GCD, WRF, and KCG performed re- search; AG analyzed data; and AG, MRS, YM, and JBF wrote the paper. landuse can either contribute to climate change *Corresponding author; e-mail: [email protected] (through degradation) or mitigate climate change

1097 1098 A. Gibbon and others

(through improved management of biosphere car- Hawaii; Weaver and Murphy 1990, Puerto Rico; bon stocks) (Denman and others 2007). The carbon Edwards and Grubb 1977, New Guinea) and also stocks and fluxes in above- and belowground from larger mountain ranges (Moser and others tropical forest components have been increasingly 2008, Ecuador; Kitayama and Aiba 2002, Borneo; studied in attempts to understand their roles in the Delaney and others 1997, Venezuela). Some site- global carbon cycle (Houghton and others 1998; specific studies have quantified carbon stock Malhi and Grace 2000) and their potential valua- changes when plantations or secondary succession tion in the newly created carbon markets and funds occurs on former cropland or pasture (Cavelier (Glenday 2006). Similarly, the size and dynamics of 1995, Columbia; Farley and others 2004, Ecuador; the above-, and, particularly, belowground carbon Fehse and others 2002, Ecuador). Likewise, some stocks of grasslands and pastures are being evalu- studies focusing on the impact of landuse on the ated as important components of the global carbon carbon stocks of high altitude grasslands have been cycle that are affected by human landuse (Scurlock conducted in the Andes (Pucheta and others 1998 and Hall 1998; Schuman and others 2002). and Adler and Morales 1999, both ). In this article, we focused our attention on an However, an integrated carbon stock quantification ecotone with poorly documented carbon stocks and study across the upper treeline of a TMCF has not high sensitivity to human and climate change yet been conducted. pressure—the puna grassland–cloud forest transi- Carbon credits can be generated through the tion of the high Andes. Tropical montane cloud voluntary carbon market for avoided deforestation, forests (TMCFs) are most generally defined as for- afforestation/reforestation and through better ests ‘‘frequently covered by cloud or mists’’ management of grasslands (Hamilton and others (Stadtmuller 1987, p. 18) and represent about 2009). However, historically, the potential to 2.5% of the world’s tropical forests (Bubb and sequester carbon in high-altitude regions has been others 2004). TMCFs are characterized by stunted, neglected, as attention has focused on lowland bi- twisting trees, laden with ephiphytes and a species omes (Fehse and others 2002). assemblage that is significantly different from the This study, from Manu National Park, Peru, tropical forests of lower altitudes (Bruijnzeel and using a combination of field and remotely sensed Veneklaas 1998). data aimed to (1) quantify, at the landscape-scale, The base of the clouds that provide TMCFs with a the aboveground carbon stocks across the upper vital moisture source are anticipated to rise due to treeline of TMCFs into the puna grasslands—an climate change, this drying coupled with rising ecotone facing pressure from climate change and temperatures makes the TMCFs a vulnerable biome landuse, (2) assess the impact of vegetation type, (Foster 2001; Still and others 1999; Bush 2002). slope, altitude, grazing and fire on the distribution The predicted rate of climate change for this cen- of carbon stocks, (3) compare results to below soil tury will require species to migrate upslope at rates carbon data gathered by Zimmerman and others an order of magnitude greater than has been (2010) at the same sample points, and (4) identify experienced in the past 50,000 years to remain in possible management options and calculate their their climatic niche (Bush and others 2004). potential effects on the carbon stocks of the area. However, a second threat to some Andean TMCFs comes from the grazing and burning of puna grasslands above, which as well as degrading the METHODS aboveground vegetation and soil of the puna Site Description (Bustamante Becerra and Bitencourt 2007) are believed to contribute to an anthropogenically Manu National Park (a World Heritage Site) ex- constrained upper limit of the treeline (Braun and tends from the eastern slopes of the Andes into the others 2002; Sarmiento and Frolich 2002; Young lowland Peruvian Amazon (IUCN 2008). The park and Leo´ n 2007). border is defined by the watershed separating the Higher elevation carbon stores are less studied catchment basins of the Urubamba and Madeira than lowland areas, but are also under threat from rivers to the south and west, and the catchment of human disturbance and could potentially benefit the Manu River in the lowlands. The National Park from an influx of carbon financing (Fehse and boundary divides ownership of the high Andean others 2002). Some studies have quantified plateau between the communities to the west and aboveground, coarse root and soil carbon up alti- the National Park to the east (Bustamante Becerra tudinal gradients of montane forests from low- and Bitencourt 2007). The study area was at the altitude island mountains (Raich and others 1997, southern extent of the Manu’s western border on Andean Treeline Carbon Stocks, Manu NP, Peru 1099

a mixture of shrubs and grasses dominated, and (3) upper TMCFs, where closed canopy trees and bam- boo dominated. Puna was predominant in the high elevation areas in the west of the study area from 4,061 m to the average treeline height of 3,450 m. The puna terrain was steeply rolling and moun- tainous, punctuated by occasional permanent and semi-permanent small lakes (up to 1 ha in area). Puna vegetation comprised grasses such as Ruiz & Pav., Calamagrostis vicunarum (Wedd.) Pilg., Festuca dolichophylla J. Presl. with an average height of approximately 30 cm. Descending in altitude, the puna was replaced by a transition zone of 0.5–2.0 m high shrubs of genera including the dominants Baccharis (Asteraceae), Hypericum (Hyp- ericaceae), Lycopodium (Lycopodiaceae), Hesperom- eles (Rosaceae), Brachyotum (Melastomataceae), and Escallonia (Escalloniaceae). The shrubs had an average height of about 1 m. The transition zone was between 1 and 50 m in width, separating the puna from cloud forest, with terrain similar to that of the puna. Below the transition zone, tree species from the Figure 1. The bottom left inset shows the location of the Asteraceae, Cunoniaceae, Melastomataceae, Rosa- study area in central Peru. The top right inset shows the full study area running northwest to southeast along ceae, Rubiaceace, and bamboo in the genus Cusquea the southeastern border of Manu National Park. Dark dominated to comprise the upper TMCFs (referred gray areas were classified as forest and white areas as puna. to hereafter as ‘‘forest’’). Patches of primary suc- The main image shows the distribution of all plots as black cession indicated disturbance from landslides and circles with a white border. The Manu National Park some areas had suffered the impacts of fire started Border is shown as a black line that delimits the western in the puna. The understory had few shrubs and extent of puna classified and goes northeast across the herbaceous plants, and trees were festooned with lowland Amazon. vascular and non-vascular epiphytes (mainly bry- ophytes, ferns, bromeliaceae, ericaceae, and or- the Paucartambo mountain range (Figure 1). The chidaceae) with a high incidence of asteraceous study area was defined as those areas above vines. Most branches were covered in bryophytes. 3,000 m inside the park boundary, extending from The forest terrain was steeply sloping with rocky the southwest boundary of the park to a northern outcrops in places. limit at 1221¢36¢¢S. An additional area of about A full description of the soil characteristics of the 3,500 ha contiguous with the SW corner of the study site can be found in Zimmerman and others park in the Wayqecha Valley was also included, (2010). Forest soils were characterized by an or- giving a surface area of the study area of approxi- ganic forest floor layer (Oh) of an average 22 cm mately 49,300 ha. with a high fine root density, overlaying a humic Rainfall averaged 1900–2500 mm annually Ah layer (11 cm) and a mineral B layer. Some across the study area, with a wet season from mineral forest soil layers contain mixed mineral October to April. Although no site-specific data layers with a high stone content as a result of an- were available, TMCFs can also receive significant cient landslides. Soil depths in the forest varied cloud water deposition inputs ranging from 0.2 to between 26 and 100 cm, with an average depth of 4mmday-1 (Bruijnzeel and Proctor 1995). Mean 44 cm. Transition zone soils were shallower and annual temperatures at 3,600 m were 11C with had only a thin Oh layer of about 2 cm, followed by frosts being common on clear nights. Various clas- an average 17-cm thick A(h) layer. Soils in the sifications for the altitudinal transition of vegeta- transition zone were between 23 and 50 cm deep tion types existed for this area (Terborgh 1977; (36 cm on average). Puna soils had no Oh layers at Young and Leo´ n 2000). For the purpose of this all and consisted mainly of organic-rich A layers study, three land-cover classes were used: (1) puna, (19 cm on average) and stony B/C layers. The where grasses dominate (2) a transition zone, where average soil profile in the puna was 33 cm deep. 1100 A. Gibbon and others

The puna had been degraded due to grazing Trees and related seasonal burning and was classified as The diameter at breast height (dbh 1.3 m above a ‘‘Zone of Recovery’’ within Manu National ground height on uphill side of stem) and height Park; being identified as an area that had been, were measured for all stems (living and dead) with and continued to be, severely damaged and a height greater than 2 m and a dbh of greater than needed special management to recover (INRENA 1 cm. Where trees did not grow straight or per- 2002). pendicular to the ground, dbh was measured 1.3 m from the base along the main stem. The tree carbon Sampling Strategy data gathered in this study are referred to as data Aboveground carbon stocks were measured across set 1 (DS1). Tree dbh and height measurements the landscape with a stratified sampling regime, from plots comprising two additional data sets col- using methods adapted from De Castro and Ka- lected within the study area were also used, these uffman (1998). Random geographical points were are referred to as DS2 and DS3 (Table 1). In total, generated for the puna or transition zone close to 5.3 ha of forest data were used for the tree biomass the treeline based on a Landsat classification. At estimation from 27 sites. each point, a transect was established that des- Allometric equations were used to estimate bio- cended 100 m in elevation into the forest and mass from the field measurements. Although spe- ascended in increments of 100 m in elevation cies-specific equations were not available for the across the puna to the highest point in the local study area, Chave and others (2005) provided an landscape for a total of 55 points; 35 in the puna, equation based on a large pan-tropical multi-spe- 8 in the transition zone, and 12 in the forest cies data set for wet climates and recommended its (Figure 1). use for montane forests (equation (1)). To compare results from other commonly used allometries, Chave and others’ (2005) dbh only, wet climate Aboveground Carbon Estimation equation (2) was used. To allow further compari- At each sample point, tree, coarse woody debris son, Brown’s (1997) dbh only, wet climate equa- (CWD), and bamboo biomass were measured in tion (3) was used to calculate aboveground tree four 3 9 15 m plots arranged around the central, biomass. Brown’s equation, like Chave and others, randomly generated point. Within each of these was based on a large pan-tropical multi-species data plots, a 1 9 5 m plot was used for shrub biomass set which was not specifically designed for mon- measurement and a 0.5 9 0.5 m plot for litter/ tane regions, and when elevation was considered, moss/grass (LMG) layer biomass determination. An the study site met Brown’s definition of ‘‘wet.’’ additional LMG plot was located at the center of the Only trees with dbh greater than 5 cm were used in arrangement. The methods used to estimate the dry the biomass estimation due to limits in the appli- biomass for each pool are described below. Dry cability of allometric equations at lower dbh values. vegetative biomass was assumed to contain 50% AGB 0:0776 q 2 0:940 carbon (after Glenday 2006). h iest ¼ ð D HÞ ð1Þ

Table 1. The Data Sets Used to Measure the Tree Carbon Stocks of the Tree Pool from Within the Study Area

Data Plot Number Total area Minimum Site Source set area (ha) of plots sampled (ha) dbh measured selection

DS1 0.018 12 0.21 1 Random at 100 m This study altitude below the treeline DS2 0.1 11 1.10 2.5 Arbitrary close W. Farfan, K. Garcia, to treeline and M. Silman1 DS3 1 4 4.00 10 Located on ridge, M. Silman with 250 m gaps in elevation Total – 27 5.31 – – –

1Unpublished. Andean Treeline Carbon Stocks, Manu NP, Peru 1101

1 hAGBiest hAGB ha iest ¼ c 6:127 ðFraction Shrub CoverÞ 2 3 ¼ q eð1:239þln 1:980ðDÞþ0:207ðlnðDÞÞ 0:0281ðlnðDÞÞ Þ ð2Þ ð5Þ where AGB_ha-1 is in Mg ha-1, correction factor 2 hAGBiest ¼ 21:297 6:953ðDÞþ0:740ðD Þð3Þ c has no units and is 2 for this site, Frac- tion_Shrub_Cover between 0 and 1. where aboveground biomass (AGB) is in kg, q Shrubs were defined as woody plants with a (wood density) is in g cm-3, D is dbh in cm, and main stem larger than 1 cm in diameter and be- H is in m. tween 0.5 and 2.0 m in height. Equation (5) was Where the wood density was available (from calculated for shrub cover in the dry desert of Chave and others’ (2006) wood density database), Argentina using three shrub species. A correction genus (63% of individuals) or family (18%) values factor of 2 was introduced to account for the mean were used. Where there was no family level data shrub height of this study (1 m), being two times available or the species was not known, the mean higher than in the data used to develop the equa- of the known trees, 0.571 g cm-3, was used. tion. This methodology would have benefitted from better calibration with the species found in the Coarse Woody Debris study area and likely represented an underestimate The length and maximum diameter of pieces of of shrub biomass for this wetter region. However, CWD within the plot were measured for all pieces shrub biomass was found to account for a small larger than 10 cm in diameter. A reduced wood fraction of overall carbon stocks in the transition density of 0.31 g cm-3 was used for all dead trees zone, suggesting uncertainty in shrub biomass had based on a mean standing deadwood density little effect on the conclusions of this study. measured in the study area. This value represented the mean value taken from unpublished data of Litter, Moss, and Grass CWD densities from two plots within the study area All litter, moss, grass, herbaceous plants, and (L.E.O. Araga˜o, unpublished data). This value shrubs with diameters smaller than 1 cm and/or concurred with Wilcke and others (2005) who also those with a diameter larger than 1 cm but height -3 recorded 0.31 g cm as an average density for all less than 50 cm were collected by clipping to CWD found above 2,000 m in a montane forest. ground level. The samples were bagged, dried at 60C for 3 days, and weighed. Bamboo Roots The dbh was measured, and height was estimated for all bamboo stems within the plot. Five bamboo Coarse root biomass (>2 mm diameter) was not shoots were measured, dried, and weighed. They directly measured due to prohibitively high costs. had a dbh range of 0.6 to 2.0 cm. A power–law Due to the effort of gathering root data, few studies relationship that related dbh to mass with R2 of were available in the literature, and none were 0.96 was then determined to be: found for the TMCFs of Peru. Therefore, Cairns and others’ (1997) commonly used, globally derived, AGB 0:2223 D 2:3264 h iest ¼ ð Þ ð4Þ mean root to shoot ratio of 0.26 was used to esti- where AGB is in kg, D is dbh in cm. mate the root biomass of the aboveground biomass Six of the bamboo shoots (<5%) measured in of trees, shrubs, and bamboo. Mokany and others the plots had a dbh greater than 2 cm. The biomass (2006) also presented 0.26 as the best linear of these shoots was capped at the biomass for regression fit in their synthesis of global forest root shoots of 2 cm dbh. to shoot ratios. A study from a TMCF in Ecuador suggested root:shoot ratios increased with altitude as a higher proportion of carbon was allocated Shrubs belowground. Leuschner and others (2007) re- Shrub biomass was calculated within each plot by ported this trend with a ratio of 0.43 at 3,060 m in modeling the crown area as an ellipse and then Ecuadorian TMCF. Thus the 0.26 value used here converting the measure to biomass per ha using the may have underestimated root carbon stocks. A linear regression equation in Flombaum and Sala standard error equal to 20% of the mean was (2007): attributed to the value. Carbon from fine root 1102 A. Gibbon and others biomass (<2 mm) was included in the soil carbon in IDRISI Andes V15. The surface area estimates of pool. the study area were improved by incorporating slope into the area calculations (Jenness, 2004). Soil The SRTM DEM was projected to 30 m resolution using the nearest neighbor method and then was Zimmerman and others (2010) quantified the carbon used to calculate the slope in each pixel in degrees. stocks of the soil at the same sample points from The surface area of each pixel was then calculated which our aboveground data were gathered by taking as the planimetric area multiplied by 1/cos(slope equal interval samples down to the bedrock. A com- angle). parison between above- and belowground stocks is Using the methodologies of Eastman (2006), presented in the ‘‘Discussion’’ section below. unsupervised and supervised (maximum likelihood Topographic and Land-Use Variables equal probability and probability weighted meth- ods) classifications of the study area were at- At each point, the maximum slope was recorded tempted. It was only possible to distinguish with a clinometer, and altitude was recorded from a between puna and forest at 30 m resolution; handheld GPS unit. Evidence of past fires (fire therefore, the narrow transition belt was incorpo- damage, charcoal in soil profile) was noted. Cattle rated into forest and puna totals. An accuracy droppings were counted within four 5 9 25 m assessment of the classifications was carried out in transects that extended out from the center of each accordance with the methodology described by random point in the puna and transition zone. Eastman (2006). The kappa index of agreement (KIA) was calculated using the error matrix module Aggregation and Extrapolation in IDRISI to assess the accuracy of the classification (Congalton 1991). A KIA of 1 indicates a perfect An analysis of the difference in carbon densities match between the classification and the observed between the three tree carbon data sets is presented. land cover, whereas 0 indicates no match (Camp- A total ecosystem carbon density (Mg C ha-1)for bell 1996). The areal extent of the land-cover class each land-cover class was estimated by summing the identified was corrected according to the results of plot carbon densities for each of the five measured the error matrix using the methodology described pools, root, and soil data. The standard error was by Dymond (1992). The final estimates of area are calculated for each carbon pool from the variance presented as the mean ± two times the root mean between plots and propagated by quadrature when square, within which the mean was ‘‘most likely to the pools were summed (for example, see Malhi and sit’’ according to Dymond (1992). others 2008, for an example of error propagation). The standard error of the total carbon content of the study site was calculated by propagating the relative RESULTS standard error of the area and the carbon densities of each land-cover class by quadrature. A t test was Land Area Classification used to test for between-biome differences if the The maximum likelihood probability weighted data were normally distributed and had similar classifier was found to be the most accurate classi- variances; otherwise a Mann–Whitney rank sum fication technique with a KIA score of 0.88. test was run (using a significance level of P = 0.05). Although no agreement on a minimum acceptable More than two groups were compared with a one- KIA exists, values greater than 0.85 are generally way ANOVA if they were normally distributed and perceived as demonstrating a ‘‘good’’ classification had similar variances. Mean values are given (Foody 2002). The maximum likelihood equal with ± 1standarderror. probability classifier and unsupervised classification Remote Sensing Classification had KIA scores of 0.872 and 0.765, respectively. According to the maximum likelihood probabil- Landsat (TM) images at 30 m spatial resolution ity weighted classifier, the study area con- were acquired for the study area (WRS-1, Path 4 / tained 29,190 ± 1,260 ha of forest and 19,480 ± Row 69). The images, acquired on the 14 August 1,060 ha of puna; 640 ± 572 ha were unidentified 2007 Shuttle Radar Topography Mission (SRTM), due to the topographic effect of shadows. The 90 m resolution Digital Elevation Model (DEM) figures presented above are ‘‘surface area,’’ as data, were received from the CGIAR consortium for opposed to ‘‘planimetric area.’’ The correction spatial information (Jarvis and others 2006). All applied for slope yielded a 15% increase in forested remote sensing and GIS processing were completed areas and a 10% increase in puna area. Forest areas Andean Treeline Carbon Stocks, Manu NP, Peru 1103

Figure 3. A comparison between the carbon stock esti- mation for each individual tree measured in the study using Chave and others’ (2005) dbh, wood density (WD) Figure 2. The land area for 100 m interval altitudinal and height, wet allometric equation (1) for tree carbon bands using the STRM DEM data and the classified and two alternative equations; Chave and others’ (2005) Landsat image. dbh and WD only, wet (equation (2)) and Brown (1997) dbh only, wet (equation (3)). were increased more due to their prevalence on increased the mean to 59 ± 4MgCha-1.Thedis- steeper slopes. Total land area in any one altitude tribution of carbon contribution from the different band was found to decline with altitude, with the dbh classes of the data sets showed that DS1 had the land area declining most above 3,600 m (Figure 2). lowest carbon stocks in every category (Figure 4). Although the average treeline height was 3,450 m, The largest differences between data sets were found puna was found at all altitudes in the study area in the above 50 cm dbh class where the DS1 mean and forest was classified up to 3,800 m. The highest of 2.3 Mg C ha-1 was significantly less than the forest site visited was 3,515 m. mean of data sets 2 and 3 with means of 11.1 and 7.2 Mg C ha-1, respectively. Differences in the Allometric Equations Equation (1) (Chave equation with height) was Carbon Density Determination for Each found to be the most conservative estimator of tree Land-Cover Class carbon (>10 cm dbh), with an equally weighed Using equation (1) and a mean tree carbon value of mean of all plots being 50 ± 6MgCha-1, whereas all sampled forest plots, the mean aboveground equation (2) (Chave without height) and equation carbon density in the forest was 63 ± 5MgCha-1 (3) (Brown) estimated 99 ± 11 and 91 ± 11 (Table 3). This was significantly (P < 0.05) greater Mg C ha-1, respectively. The biomass estimates for than the carbon densities of the transition zone individual trees from the three equations diverged 16.9 ± 2.2 Mg C ha-1 and puna 7.5 ± 0.7 Mg C with increasing dbh (Figure 3). Equation (1) pre- ha-1. The forest’s aboveground carbon stock was dicted consistently lower values than equations (2) dominated by trees with more carbon found in the and (3) and was used for all subsequent calculations. litter/moss/grass layer than in the CWD pool. The Variation Between the Tree Carbon Pool LMG layer contained the largest fraction of aboveground carbon in both the transition zone Data Sets and puna. The transition zone had more than When the three tree carbon pool data sets (>10 cm double the carbon in the LMG layer than the puna, dbh only) were considered separately, DS1’s carbon which accounts for the majority of the difference density of 35 ± 9MgCha-1 was lower than DS2’s between the two. The shrubs, although a dominant and DS3’s with 63 ± 8 and 59 ± 6MgCha-1, visual feature of the transition zone, contributed respectively (Table 2). The mean tree carbon relatively little to total aboveground biomass, even density, with all data points considered equally, if the allometric relationship we employed under- was 50 ± 6MgCha-1, weighting by plot area estimated biomass by a factor of three. 1104 A. Gibbon and others

Table 2. Comparison of Tree Carbon Pool Data from the Four Data Sets used for Trees Larger than 10 cm Only

Data Altitudinal Mean tree carbon Mean tree Mean dbh (cm) range density ± SE (Mg C ha-1) height (m)

DS1 3000–3515 35 ± 9 7.7 18.5 DS2 3350–3630 63 ± 8 8.8 20.3 DS3 3000–3450 59 ± 6 9.3 20.0 All data 3000–3630 54 ± 3 9.2 20.0

Belowground carbon stocks were seen to represent an increasing proportion of the total stock as alti- tude increased, again due to the transition to puna with a higher belowground:aboveground carbon stock ratio (Figure 5). When Zimmerman and others’ data are considered, in total 75% of the carbon in the study area was stored belowground in roots and soil. Given that there was no rela- tionship between slope and altitude, the total aboveground carbon stock was calculated by scal- ing the mean carbon densities of puna and forest by the spatial area covered when corrected for slope at 30 m resolution. The total carbon stock 8.2 ± 0.6 Tg comprised 2.3 ± 0.2 Tg from the puna and 5.9 ± 0.5 Tg from the forest.

DISCUSSION Figure 4. The contribution of dbh classes to the total The Effect of Topography and Land-Use carbon stock of plots from the three data sets used in this study. Variables Aboveground biomass was not significantly related The Total Carbon Stock of the Study Area to slope, topographical position, or altitude within any of the land-cover classes, with the exception of The carbon stock of the study area gradually de- the positive correlation between aboveground car- creased from 3,000 m to 3,600 m (Figure 5) as both bon density and altitude of the transition zone land area decreases (Figure 2) and the prevalence (P < 0.05). This correlation could be due to a third of puna relative to forest increased steadily.

Table 3. Carbon Densities for Puna, Transition Zone, and Forest

Carbon pool Mean carbon density (Mg C ha-1 ± 1 SE)

Puna Transition zone Forest (N = 35) (N =8) (N = 27 for trees, 12 for others)

Litter/moss/grass 6.8 ± 0.6 13.3 ± 1.5 6.6 ± 0.8 Shrubs 0.5 ± 0.2 2.6 ± 1.5 0.5 ± 0.3 CWD 0 0 1.8 ± 0.9 Bamboo 0.03 ± 0.28 0.14 ± 0.07 0.9 ± 0.4 Trees (>10 cm dbh) 0.1 ± 0.1 0.8 ± 0.7 50 ± 5 Trees (>5 cm and <10 cm dbh) 0.1 ± 0.1 0.1 ± 0.1 3.5 ± 0.7 Total above ground 7.5 ± 0.7 16.9 ± 2.3 63.4 ± 5.2 Roots 0.05 ± 0.01 0.23 ± 0.04 13.9 ± 2.8 Soil carbon 119 ± 8 147 ± 14 118 ± 15 Total carbon 127 ± 8 164 ± 14 195 ± 16 Andean Treeline Carbon Stocks, Manu NP, Peru 1105

Figure 5. The change in total carbon stock for 100 m altitudinal bands divided by land-cover type (left) and between aboveground (AG) and belowground (soil and roots; right) for the study area.

factor; higher treelines existed in places that were (P > 0.15). Evidence of past fires was found at less affected by fire. 26% of puna sites and in 25% of the forest sites Of the 27 puna points where grazing was ana- from DS1. Our sampling did not find a significant lyzed, 69% had evidence of grazing (at least 1 difference in the aboveground carbon densities of dropping found), whereas one of the five transition those sites with historical evidence of fire and those points assessed was grazed (Table 4). Puna points without (within the same land-cover class). Our with evidence of grazing had a mean aboveground methodology would not capture data related to carbon density of 6.5 ± 0.3 Mg C ha-1, signifi- carbon losses due to the transition from a higher- cantly different from 8.9 ± 0.6 Mg C ha-1 for to-lower carbon density land-cover class. those without evidence (P < 0.05). There was a Puna sites with evidence of past fires were found to negative relationship between the cattle dropping have a mean soil carbon density of 96 ± 11 Mg count and the aboveground carbon density of Cha-1 compared to 123 ± 11 Mg C ha-1 for those puna; however, the relationship was not significant without (Zimmerman and others 2010). Those sites with evidence of grazing had a mean soil carbon density of 111 ± 10 Mg C ha-1 compared to 124 ± Table 4. Topographical and Landuse Results 14 Mg C ha-1 without (Zimmerman and others from the Study Area 2010). However, there was not a statistically signif- Land cover Puna Transition Forest icant difference between either of these relation- zone ships. Further analysis of the Zimmerman and others data revealed that, when only the top 10 cm of soil Total sites 35 8 27 was analyzed, the soil carbon density in grazed sites Altitude of 56 ± 2MgCha-1 was significantly higher than Mean 3,540 3,450 3,345 -1 Min 3,350 3,280 3,000 in ungrazed areas 48 ± 2MgCha (P < 0.05). Max 3,860 3,620 3,630 This appeared to be a compaction effect because Maximum slope there was a corresponding significant increase in -3 Mean 37 45 68 bulk density of grazed sites (0.55 ± 0.02 g cm vs. Min 7 15 0 0.45 ± 0.02 g cm-3, P < 0.05), yet no significant Max 110 71 95 difference in carbon concentrations (both 14 ± 1 Grazing %). Forest sites affected by fire had higher mean Much 6 0 NA carbon stocks 148 ± 25 Mg C ha-1 compared to Some 14 1 NA non-affected sites 120 ± 40 Mg C ha-1; however, None 9 4 NA this difference was not significant, primarily due to Not recorded 6 3 NA the small sample size and large variation between Fire Fire evidence 9 0 3 sites (Zimmerman and others 2010). No evidence 26 8 9 Comparisons with Other Studies Twenty-nine sites were used for tree carbon density measurement, but 12 (DS1) were sampled for coarse woody debris, bamboo, shrub and LMG pools, maximum Zimmerman and others (2010) found that soil slope and for historical evidence of fire. carbon densities were not significantly different 1106 A. Gibbon and others between forest, transition zone, and puna the Argentinean montane grasslands studied by (118 ± 15, 147 ± 14, and 119 ± 8MgCha-1, Pucheta and others (1998). respectively). Using Zimmerman and others’ data, The lack of slope or altitudinal effects on carbon the ratio of belowground:aboveground carbon (all densities for either the puna or forest sites sug- pools) decreased from 15.8 on the puna to 8.6 in gested other factors such as micro-climate or the transition zone and 2.1 in the forest. There landuse (both current and historical) were more were no significant relationships between above- important factors in stock distribution. Fire is ground and belowground carbon densities within known to have an immediate impact of reducing land-cover classes. carbon stocks through combustion (Van Der Werf The carbon density of trees larger than 5 cm dbh and others 2003). Although the results presented (54 ± 6MgCha-1) in this study was comparable here did not show any relationship between past with those of Moser and others (2008), who re- fires and reduced carbon stocks, our sampling did corded 56.1 Mg C ha-1 for all trees with dbh above not aim to thoroughly capture the range of fire 3 cm in a site selected as an example of closed impact experienced by the land-cover classes, canopy upper montane forest at 3,060 m in Ecua- which would be necessary to draw conclusions on dor. Secondary successional forest stands, of Alnus the effects of fire. acuminata (Betulaceae) and Polylepis incana (Rosa- ceae) at 3,200 m in Ecuador where upper TMCF had previously been deforested, had higher carbon Uncertainties densities of 120 and 183 Mg C ha-1, respectively Uncertainties can be classified as either random or (Fehse and others 2002). In Borneo, Kitayama and systematic. Random uncertainties, caused by spa- Aiba (2002) report tree carbon densities of 18 and tial variation, can largely be addressed through an 107 Mg C ha-1 from two plots in ‘‘Subalpine forest appropriate random (or random stratified) sam- /scrub’’ (3,100 m) as well as 60 and 150 Mg C ha-1 pling protocol, and are appropriately quantified for from two plots in ‘‘Upper Montane Rainforest’’ our soil, litter layer, and dead wood sampling. (2,700 m). Chave and others (2004) found that when extrap- The wide range of carbon values found in the olating forest inventory measurement to the land- literature accurately reflected the range of carbon scape scale, the largest contributors to uncertainty densities that were found in forested high elevation were the choice of allometric equation, the size of landscapes. The landscape average, however, re- study plots, and their representativeness of plots in flected myriad factors, including succession after the landscape. disturbances, environmental factors such as depth The choice of allometric equation significantly to water table and local orographic effects and their affected the aboveground carbon results. Equation relative abundances across the landscape. Extrap- (1), which used tree height and dbh, gave signifi- olations to the landscape, a central goal of ecosys- cantly lower carbon estimates of the tree carbon tem ecology and required for use in ecosystem pool than the other two equations which used dbh services markets, must weigh the relative impor- only. The estimation of tree heights in TMCFs tance of these factors across the landscape. We used where trees do not always grow straight was diffi- a random sampling approach, which is labor cult. However, Williams and Schreuder (2000) intensive but eliminated many sources of bias due conclude that, as long as height estimates are cor- to site selection. Further work should combine rect ±40%, they still improve allometric equation randomized sampling schemes with remote sensing accuracy. Systematic errors in height measurement efforts to produce accurate landscape-level carbon that would explain the significant difference in stock estimates for high elevation systems. results from the with- and without-height equa- Grazed puna sites which had 6.5 ± 0.3 Mg tions are unlikely. Thus, it was thought the differ- Cha-1 in aboveground carbon densities were ence in results was due to the sample of trees used comparable with the regularly burned and grazed to generate equations (2)and(3) having a different Ecuadorian paramo grasslands with aboveground dbh to height relationship for use in these upper densities of 4.0–4.2 Mg C ha-1 found at 3,750 m TMCFs. Although no testing of the allometric and 4,000 m sites, respectively (Ramsay and Oxley equations with actual tree biomass data was possi- 2001). The grazed sites had an aboveground carbon ble, equation (1) was the most accurate equation in density 27% lower than the ungrazed sites, similar a study by Moser and others (2008) when com- to the difference of 33% found between grazed pared to biomass data available from three wind sites and those excluded from grazing for 2 years in thrown trees in an Ecuadorian TMCF. Andean Treeline Carbon Stocks, Manu NP, Peru 1107

Chave and others (2004) recommended at least Potential to Increase Carbon Stocks 5 ha of sampling for a landscape scale assessment Here, we quantified the potential volume of carbon of forest carbon to ensure a representative sample sequestration resulting from projects to reduce of the landscape was taken. This study used a combination of large (1 ha) plots (DS3) and degradation of puna from overgrazing and facili- smaller plots (DS1 and DS2), that totaled 5.3 ha. tating the upslope migration of the treeline in re- sponse to climate change. Conservative carbon DS1, which had the smallest and only completely sequestration totals under the scenarios were cal- random distribution of plots recorded the lowest culated by subtracting the current carbon densities carbon densities. The largest difference between plus one standard error from the expected carbon the carbon content of the DS1 forest plots and density minus one standard error. the others was in the larger than 50 cm dbh class of trees. This could have been because sites for DS2 and DS3 were non-random and were se- Controlled Grazing lected as good examples of forest, avoiding bam- boo patches and large clearings resulting from The results presented here are consistent with landslides. Another reason could have been the evidence that over-grazing of high altitude grass- non-normal distribution of large but rare trees lands can reduce aboveground carbon densities (Chave and others 2003), which a small number (Pucheta and others 1998; Adler and Morales of small sampling plots below the recommended 1999). Therefore, improved grazing management minimum plot size of 0.25 ha were likely to miss. schemes in such areas could lead to carbon Within each data set and considering all data sequestration benefits. together, no significant relationship was found If the percentage of grazed sites (67%) was rep- between the distance from treeline or altitude resentative of the extent of grazed land within the and total carbon density. Although a decrease in study area, then approximately 13,000 ha were total carbon density of forest may have been affected by grazing. If the mean carbon densities of expected at the treeline approaches (Moser and these grazed areas were raised to the non-grazed others 2008), the narrow altitudinal range of mean values then with conservative gains of -1 plots (3,000–3,625 m) and observed differences in 1.5 Mg C ha aboveground 0.019 Tg C would be treeline forest structure (from gradual thinning sequestered. Due to the variation between sites, a through a transition zone to abrupt mature for- more intense and focused study on the effects of est–puna transitions) explain the lack of correla- grazing on soil carbon in the study area is needed tion. Therefore, results from DS1 were likely to before any significant trends can be identified. be an underestimate of the true mean carbon density of the forest at the treeline. With the Increasing Forest Cover possibility that results from DS2 and DS3 were overestimates, a mean of DS1-3 was considered Barriers to succession of forests onto pastures in- an appropriate best estimation. clude the impact of grazing (Cavelier 1995), com- The methodology was designed to capture all petition with dense grasses, and the lack of seed significant carbon stocks in the landscape; how- dispersal (Holl and others 2000). Furthermore, the ever, some small pools were not quantified. A rate of TMCF’s natural succession onto pastures is study of elfin forest in Ecuador (Moser and low, even when a source of seed dispersal was others 2008) found that trees 3–5 cm dbh con- close, with a canopy dominated by shade tolerant tributed 9% of biomass. However using equation species possibly taking several hundred years to (1) beyond its recommended range of dbh values establish (Golicher and Newton 2007). Therefore, to include trees 2–4.99 cm dbh for DS1 and DS2, to facilitate the upslope migration of TMCF species, we found they contributed less than 2% of the assisted afforestation through seeding, or planting total above ground biomass (<1MgCha-1). shrubs or native tree seedlings could be carried out Epiphytes such as lichens, mosses, and bromel- (see Holl and others 2000). Projects with a primary iads were not counted but were not expected to objective of carbon sequestration and revenue contribute significantly to carbon stocks (Edwards generation such as monoculture plantations of Pi- and Grubb 1977). It was noted that the land- nus or Eucalyptus species have questionable benefits scape contained sporadic lakes and bogs smaller for biodiversity and water supplies (Cavelier 1995; than 1 ha in size, whose belowground carbon Hofstede and others 2002) and thus regeneration stocks could be significant but required more projects are considered more appropriate for pro- sampling. tected areas such as Manu National Park. 1108 A. Gibbon and others

2007). Soil carbon stocks of puna and forest were found to be equal; therefore, with simple assump- tions that these stocks were in equilibrium, and a new equilibrium would be reached after afforesta- tion with native species, no net change in soil carbon would be expected. However, the distribu- tion of carbon through the soil profile was not the same in forest and puna soils, with forests being characterized by a 22-cm Oh layer and an 11-cm Ah layer, whereas the puna soils had only a 19-cm Ah layer. Therefore, under an optimistic scenario, with no losses to the current puna Ah layer and an accumulation of 20 cm of Oh layer, carbon stocks could increase by 65 Mg C ha-1 (data from Zim- mermann and others 2010). When added to the 77.3 Mg C ha-1 that could be stored in above- ground and coarse root biomass, a total 2.3 Tg C could be sequestered in the 49,300-ha study area (Figure 6). The assumption of no change in soil carbon Figure 6. Estimated maximum potential gains in stocks under an afforestation scenario was sup- aboveground (AG) and root carbon under an afforesta- ported by Paul and others (2002) who found pas- tion of puna scenario using the mean carbon densities ture to forests conversions experienced a loss in soil found in this study (black) on top of current stocks (white). And the estimated gains in soil carbon based on carbon over the first 5 years but a recovery to pre- an assumption of no change in the existing puna soil afforestation levels after 30 years. Evidence from carbon stock and a gain of 20 cm of Oh layer (striped) meta-analyses showed no significant change in soil using a mean of the carbon stock in the 0–20 cm layer of carbon when broad leaf forests were established or forest soils in the study area. secondary succession occured on pasture (Guo and Gifford 2002). Del Castillo and Blanco-Macı´as (2007) found after 100 years, abandoned cropland Raising the mean puna aboveground and root under secondary succession of TMCF accumulated carbon densities to that of the forest would -1 an O horizon (including litter) of 15–30 cm, sug- sequester 62.4 Mg C ha . Afforesting the entire gesting gains in Oh carbon stocks could be achieved puna area would sequester 1.1 Tg C, however, in at least this amount of time. 60% of these gains could be made with afforesta- Given the relative size of the carbon stock in high tion below 3,600 m (Figure 6). Rates of TMCF elevation pastures, best practice management that forest recovery are known to be relatively low reduces carbon loss from soils should be executed (Golicher and Newton 2007) and with a space-by- (Paul and others 2002). High monitoring costs may time substitution experiment on ex-cropland at discourage monitoring soil carbon in afforestation 1850 m in Mexico, Del Castillo and Blanco-Macı´as schemes on high altitude pastures (Robertson and (2007) were able to show secondary succession of others 2004); however, they may be necessary for TMCF took up to 100 years to reach maturity and truly conservative carbon stock change estimates maximum carbon stocks. Strategies to speed-up the and may reveal positive gains if the loss of Ah layer rate of forest regeneration should be central to any carbon is less than that of Oh layer accumulation. management effort, such as planting rapidly growing nurse trees that are attractive to avian seed dispersers. CONCLUSION Carbon sequestration projects now have the poten- The Importance of the Soil Carbon Stock tial to earn financing through sales on the voluntary Given the size of the carbon stock in the puna soils and regulated carbon markets. The sequestration of relative to the aboveground stock of the forest 1.5 Mg C ha-1 across the 18,000 ha of puna in the (about twice as large), the fate of this stock under study area through improved grazing management, an afforestation scenario has important implica- assuming a carbon price of US $3.4 per t CO2e (the tions for the net carbon budget of the area (Jackson average price for agricultural carbon projects in and others 2002; Gonza´lez-Espinosa and others 2008, Hamilton and others 2009) would earn only Andean Treeline Carbon Stocks, Manu NP, Peru 1109

USD 65,000. This is very unlikely to cover project biodiversity—a global assessment. London: Parthenon Pub- development, monitoring, and verification costs. lishing Group. p 75–89. However, assisted migration of the treeline may not Brown S. 1997. Estimating biomass and biomass change of tropical forests: a primer. FAO Forestry Paper No. 134, 55 pp be possible without relieving grazing and fire pres- Bruijnzeel LA, Proctor J. 1995. Hydrology and biogeochemistry sure on the treeline. If only the aboveground and of tropical montane cloud forests: what do we really know? coarse root carbon stocks were considered, and the In: Hamilton LS, Juvik JO, Scatena FN, Eds. Tropical montane average carbon price for afforestation conservation cloud forests. Ecological Studies, vol 110. New York: Springer. projects of US $7.5 per t CO2e were achieved (Ham- pp. 38–78. ilton and others 2009), this would represent an in- Bruijnzeel LA, Veneklaas EJ. 1998. Climatic conditions and come of US$ 1,374 per ha over the project lifetime for tropical montane forest productivity: the fog has not lifted yet. Ecology 79:3–9. afforestation of puna. If soil carbon gains occurred Bubb P, May I, Miles L, Sayer J. 2004. Cloud forest agenda. and were monitored, this number could increase by Cambridge: UNEP-WCMC. p 34. up to a factor of 2. Bush MB. 2002. Distributional change and conservation on the There are still considerable uncertainties over the Andean flank: a palaeoecological perspective. Glob Ecol Bio- feasibility of such a project including whether suf- geogr 11:463–73. ficient management capacity exists within the park Bush MB, Silman MR, Urrego DH. 2004. 48,000 years of climate to implement such a project; whether the local and forest change in a biodiversity hot spot. Science communities could be sufficiently engaged and 303:827–9. suitably compensated to ensure the project’s suc- Bustamante Becerra JA, Bitencourt MD. 2007. Ecological zoning of an Andean grasslands (puna) at the manu biosphere re- cess. Likewise the impacts on water yield in the serve, Peru. Int J Environ Sustain Dev 6:357–72. catchment through altered transpiration, intercep- Buytaert W, In˜ iguez V, Bie`vre BD. 2007. The effects of affores- tion, evaporation, and cloud water deposition re- tation and cultivation on water yield in the Andean pa´ramo. quire careful consideration (Buytaert and others For Ecol Manag 251:22–30. 2007; Farley and others 2005). However, Manu Cairns MA, Brown S, Helmer EH, Baumgardner GA. 1997. Root National Park presents a real opportunity where biomass allocation in the world’s upland forests. Oecologia assisted migration of the treeline through affores- 111:1–11. tation combined with grazing management could Campbell JB. 1996. Introduction to remote sensing. 2nd edn. New York (NY): Guilford. both increase carbon stocks and increase the Cavelier J. 1995. Reforestation with the native tree Alnus ac- chances of endemic species survival in the face of uminata: effects on phytodiversity and species richness in an climate change. upper montane rain forest area of Colombia. In: Hamilton LS, Juvik JO, Scatena FN, Eds. Tropical montane cloud forests. ACKNOWLEDGMENTS New York (NY): Springer. p 125–37. Chave J, Condit R, Lao S, Caspersen JP, Foster RB, Hubbell SP. We thank the Blue Moon Fund for support. We 2003. Spatial and temporal variation of biomass in a tropical especially thank Manu National Park and the forest: results from a large census plot in Panama. J Ecol Peruvian Instituto Nacional de Recusos National 91:240–52. (INRENA) and the Amazon Conservation Associa- Chave J, Condit R, Aguilar S, Hernandez A, Lao S, Perez R. 2004. tion (ACCA) for permission to work in Manu Na- Error propagation and scaling for tropical forest biomass esti- mates. Philos Trans R Soc B 359:409–20. tional Park and at the Wayquecha field station, Chave J, Andalo C, Brown S, Cairns MA, Chambers JQ, Eamus respectively. The guards at Manu National Park and D, Fo¨ lster H, Fromard F, Higuchi N, Kira T. 2005. Tree Luis Imunda provided essential logistical support allometry and improved estimation of carbon stocks and bal- and advice. Five hard-working undergraduate stu- ance in tropical forests. Oecologia 145:87–99. dents from Wake Forest University and Flor Za- Chave J, Muller-Landau HC, Baker TR, Easdale TA, Steege H, mora, Percy Chambi and Nelson Cahuana from Webb CO. 2006. Regional and phylogenetic variation of wood Universidad Nacional de San Antonio Abad del density across 2456 neotropical tree species. Ecol Appl 16:2356–67. Cusco, were essential for the completion of this Congalton RG. 1991. A review of assessing the accuracy of project. classifications of remotely sensed data. Remote Sens Environ 37:35–46. REFERENCES De Castro EA, Kauffman JB. 1998. Ecosystem structure in the Brazilian Cerrado: a vegetation gradient of aboveground bio- Adler PB, Morales JM. 1999. Influence of environmental factors mass root mass and consumption by fire. J Trop Ecol and sheep grazing on an Andean grassland. J Range Manag 14:263–83. 52:471–81. Del Castillo RF, Blanco-Macı´as A. 2007. Secondary succession Braun G, Mutke J, Reder A, Bartbloft W. 2002. Biotope patterns, under a slash-and-burn regime in a tropical montane cloud phytodiversity and forestline in the Andes, based on GIS and forest: soil and vegetation characteristics. In: Newton AC, Ed. remote sensing data. In: Korner C, Spehn E, Eds. Mountain Biodiversity loss and conservation in fragmented forest land- 1110 A. Gibbon and others

scapes: the forests of montane Mexico and temperate South Guo LB, Gifford RM. 2002. Soil carbon stocks and land use America. Wallingford: CABI Publishing. p 158–80. change: a meta analysis. Glob Change Biol 8:345–60. Delaney M, Brown S, Lugo AE, Torres-Lezama A, Bello Quintero Hamilton K, Sjardin M, Shapiro A, Marcello T. 2009. Fortifying N. 1997. The distribution of organic carbon in major compo- the foundation: state of the voluntary carbon markets 2009. nents of forests located in five life zones of Venezuela. J Trop Ecosystem Marketplace and New Carbon Finance. Ecol 13:697–708. Hofstede RGM, Groenendijk JP, Coppus R, Fehse JC, Sevink J. Denman KL, Brasseur G, Chidthaisong A, Ciais P, Cox PM, 2002. Impact of pine plantations on soils and vegetation in the Dickinson RE, Hauglustaine D, Heinze C, Holland E, Jacob D, Ecuadorian High Andes. Mt Res Dev 22:159–67. Lohmann U, Ramachandran S, da Silva Dias PL, Wofsy SC, Holl KD, Loik ME, Lin EHV, Samuels IA. 2000. Tropical montane Zhang X. 2007. Couplings between changes in the climate forest restoration in Costa Rica: overcoming barriers to dis- system and biogeochemistry. In: Solomon S, Qin D, Manning persal and establishment. Restor Ecol 8:339–49. M, Chen Z, Marquis M, Averyt KB, Tignorand M, Miller HL, Houghton RA, Davidson EA, Woodwell GM. 1998. Missing sinks Eds. Climate change 2007: the physical science basis contri- feedbacks and understanding the role of terrestrial ecosystems bution of working group I to the fourth assessment report of in the global carbon balance. Glob Biogeochem Cycles the intergovernmental panel on climate change. New York, 12:25–34. Cambridge: Cambridge University Press. INRENA. 2002. Instituto Nacional de Recursos Naturales Plan Dymond JR. 1992. How accurately do image classifiers estimate Maestro—Proyecto Aprovechamiento y Manejo Sostenible de area? Int J Remote Sens 13:1735–42. la Reserva de Biosfera y Parque Nacional del Manu (Pro- Eastman J. 2006. IDRISI Andes—guide to GIS and image pro- Manu) Peru: INRENA. cessing. Clark Labs [online]. http://planetuwcacza/nisl/ IUCN. 2008. Protected areas and world heritage Peru—Manu Gwen’s%20Files/GeoCourse/Resource%20Mapping/Andes% National Park. UNESCO IUCN WHC [online] UNEP. http:// 20Manualpdf. Accessed 14 Feb 2008. wwwunep-wcmcorg/sites/wh/pdf/Manupdf. Accessed 08 Jan Edwards PJ, Grubb PJ. 1977. Studies of mineral cycling in a 2009. montane rain forest in New Guinea I: the distribution of or- Jackson RB, Banner JL, Jobbagy EG, Pockman WT, Wall DH. ganic matter in the vegetation and soil. J Ecol 65:943–69. 2002. Ecosystem carbon loss with woody plant invasion. Farley KA, Kelly EF, Hofstede RGM. 2004. Soil organic carbon Nature 418:623–6. and water retention after conversion of grasslands to pine Jarvis A, Reuter H, Nelson A, Guevara E. 2006. Hole-filled plantations in the Ecuadorian Andes. Ecosystems 7:729–39. seamless SRTM data v3. Cali: International Center for Tropical Farley KA, Jobbagy EG, Jackson RB. 2005. Effects of afforesta- Agriculture. tion on water yield: a global synthesis with implications for Jenness JS. 2004. Calculating landscape surface area from digital policy. Glob Change Biol 11:1565–76. elevation models. Wildl Soc Bull 32:829–39. Fehse J, Hofstede R, Aguirre N, Paladines C, Kooijman A, Sevink Kitayama K, Aiba SI. 2002. Ecosystem structure and productivity J. 2002. High altitude tropical secondary forests: a competitive of tropical rain forests along altitudinal gradients with con- carbon sink? For Ecol Manag 163:9–25. trasting soil phosphorus pools on Mount Kinabalu Borneo. Flombaum P, Sala OE. 2007. A non-destructive and rapid J Ecol 90:37–51. method to estimate biomass and aboveground net primary Leuschner C, Moser G, Bertsch C, Ro¨ derstein M, Hertel D. 2007. production in arid environments. J Arid Environ 69:352–8. Large altitudinal increase in tree root/shoot ratio in tropical Foody G. 2002. Status of land cover classification accuracy mountain forests of Ecuador. Basic Appl Ecol 8:219–30. assessment. Remote Sens Environ 80:185–201. Malhi Y, Grace J. 2000. Tropical forests and atmospheric carbon Foster P. 2001. The potential negative impacts of global climate dioxide. Trends Ecol Evol 15:332–7. change on tropical montane cloud forests. Earth-Sci Rev Malhi Y, Araga˜o LEO, Metcalfe DB, Paiva R, Quesada CA, Al- 55:73–106. meida S, Anderson L, Brando P, Chambers JQ, da Costa ACL, Glenday J. 2006. Carbon storage and emissions offset potential Hutyra LR, Oliveira P, Patin˜ o S, Pyle EH, Robertson AL, Te- in an East African . For Ecol Manag ixeira LM. 2008. Comprehensive assessment of carbon pro- 235:72–83. ductivity allocation and storage in three Amazonian forests. Golicher D, Newton AC. 2007. Applying succession models to Glob Change Biol 15:1255–74. the conservation of tropical montane forest in biodiversity loss Mokany K, Raison J, Prokushkin A. 2006. Critical analysis of and conservation. In: Newton AC, Ed. Fragmented forest root:shoot ratios in terrestrial biomes. Glob Change Biol landscapes: the forests of montane Mexico and temperate 12:84–96. . Wallingford: CABI Publishing. p 200–20. Moser G, Roderstein M, Soethe N, Hertel D, Leuschner C. 2008. Gonza´lez-Espinosa M, Ramı´rez-Marcial N, Newton A, Rey- Altitudinal changes in stand structure and biomass allocation Benayas JM, Camacho-Cruz A, Armesto JJ, Lara A, Premoli of tropical mountain forests in relation to microclimate and AC, Williams-Linera G, Altamirano A, Alvarez-Aquino C, soil chemistry. Ecol Stud 198:229–42. ´ Cortes M, Galindo-Jaimes L, Mun˜ iz MA, Nu´n˜ ez M, Pedraza Paul KI, Polglase PJ, Nyakuengama JG, Khanna PK. 2002. RA, Rovere AE, Smith-Ramı´rez C, Thiers O, Zamorano C. Change in soil carbon following afforestation. For Ecol Manag 2007. Restoration of forest ecosystems in fragmented land- 168:241–57. scapes of temperate and montane tropical Latin America. In: Newton AC, Ed. Biodiversity loss and conservation in frag- Pucheta E, Cabido M, Dı´az S, Funes G. 1998. Floristic compo- mented forest landscapes: the forests of montane Mexico and sition biomass and aboveground net plant production in temperate South America. Wallingford: CABI Publishing. grazed and protected sites in a mountain grassland of central Argentina. Acta Oecol 19:97–105. p 355–69. Andean Treeline Carbon Stocks, Manu NP, Peru 1111

Raich JW, Russell AE, Vitousek PM. 1997. Primary productivity Van Der Werf GR, Randerson JT, Collatz GJ, Giglio L. 2003. and ecosystem development along an elevational gradient on Carbon emissions from fires in tropical and subtropical eco- Mauna Loa Hawaii. Ecology 78:707–21. systems. Glob Change Biol 9:547–62. Ramsay PM, Oxley ERB. 2001. An assessment of aboveground Weaver PL, Murphy PG. 1990. Forest structure and productivity net primary productivity in Andean grasslands of Central in Puerto Rico’s Luquillo Mountains. Biotropica 22:69–82. Ecuador. Mt Res Dev 21:161–7. Wilcke W, Hess T, Bengel C, Homeier J, Valarezo C, Zech W. Robertson K, Loza-Balbuena I, Ford-Robertson J. 2004. Moni- 2005. Coarse woody debris in a montane forest in Ecuador: toring and economic factors affecting the economic viability of mass C and nutrient stock and turnover. For Ecol Manag afforestation for carbon sequestration projects. Environ Sci 205:139–47. Policy 7:465–75. Williams MS, Schreuder HT. 2000. Guidelines for choosing Sarmiento FO, Frolich LM. 2002. Andean cloud forest tree lines: volume equations in the presence of measurement error in naturalness agriculture and the human dimension. Mt Res height. Can J For Res 30:306–10. Dev 22:278–87. Young KR, Leo´ n B. 2000. Biodiversity conservation in Peru’s Schuman GE, Janzen HH, Herrick JE. 2002. Soil carbon eastern montane forests. Mt Res Dev 20:208–11. dynamics and potential carbon sequestration by rangelands. Young KR, Leo´ n B. 2007. Tree-line changes along the Andes: Environ Pollut 116:391–6. implications of spatial patterns and dynamics. Philos Trans R Scurlock JMO, Hall DO. 1998. The global carbon sink: a grass- Soc B 362:263–72. land perspective. Glob Change Biol 4:229–33. Zimmermann M, Meir P, Silman MR, Fedders A, Gibbon A, Stadtmuller T. 1987. Cloud forests in the humid tropics—a bib- Malhi Y, Urrego DH, Bush MB, Feeley KJ, Garcia KC, Dargie liographic review. Turrialba: CATIE. GC, Farfan WR, Goetz BP, Johnson WT, Kline KM, Modi AT, Still CJ, Foster PN, Schneider SH. 1999. Simulating the effects of Rurau NMQ, Staudt BT, Zamora F. 2010. No differences in soil climate change on tropical montane cloud forests. Nature carbon stocks across the in the Peruvian Andes. 398:608–10. Ecosystems 13:62–74. Terborgh J. 1977. Bird species diversity on an Andean eleva- tional gradient. Ecology 58:1007–19.